Keras ImagedataGenerator Documenation

Documentation: https://keras.io/preprocessing/image/

keras.preprocessing.image.ImageDataGenerator(featurewise_center=False,
samplewise_center=False,
featurewise_std_normalization=False,
samplewise_std_normalization=False,
zca_whitening=False,
zca_epsilon=1e-6,
rotation_range=0.,
width_shift_range=0.,
height_shift_range=0.,
shear_range=0.,
zoom_range=0.,
channel_shift_range=0.,
fill_mode='nearest',
cval=0.,
horizontal_flip=False,
vertical_flip=False,
rescale=None,
preprocessing_function=None,
data_format=K.image_data_format())

ImageGenerator_Visualizer

Example

In [1]:
%matplotlib notebook
from ImageGenerator_Visualizer import *

Using TensorFlow backend.
In [2]:
%%javascript
IPython.OutputArea.prototype._should_scroll = function(lines) {
    return false;
}
In [3]:
PATH = './images/'
images = sorted(glob.glob(PATH+'*'))
img_sz = (432,292)
imgs = np.stack([np.array(Image.open(img).convert('RGB').crop((0,0,432,292))) for img in images])
In [4]:
parameter_space = {}

parameter_space['featurewise_center'] = [True]
parameter_space['samplewise_center'] = [True]
parameter_space['featurewise_std_normalization'] = [True]
parameter_space['samplewise_std_normalization'] = [True]
parameter_space['rotation_range'] = [10, 20, 50, 360]
parameter_space['width_shift_range'] = [0.05, 0.1, 0.2, 0.5]
parameter_space['height_shift_range'] = [0.05, 0.1, 0.2, 0.5]
parameter_space['shear_range'] = [0.05, 0.1, 0.2, 0.5]
parameter_space['zoom_range'] = [0.05, 0.1, 0.2, 0.5]
parameter_space['channel_shift_range'] = [0.05, 0.1, 0.2, 0.5]
parameter_space['fill_mode'] = ['constant', 'nearest', 'reflect', 'wrap']
parameter_space['cval'] = [0.05, 0.1, 0.2, 0.5]
parameter_space['horizontal_flip'] = [True]
parameter_space['vertical_flip'] = [True]

Note:

  • fill_mode is only applicable if there is empty space, which requires to be filled (e.g. rotation, shift, zoomout). The visualization function sets height_shift_range = 0.5
  • cval is only applicable for fill_mode = 'constant'. The visualization function sets fill_mode = 'constant' and height_shift_range = 0.5
In [5]:
visualization(imgs, parameter_space, './output_images/')